FOR AUTISTIC ADULTS & ADULTS WITH IDD · ALWAYS FREE · PART OF THE ALWAYSHERE FAMILY
A guardian inside their chat.
This is NOT another chatbot. Guardian is a safety and personalization layer inside the AI your loved one already uses. It learns how they communicate and what they need, and it texts you when something isn't right — any hour they choose to chat, 24/7/365.
Always free. No card. Nothing new for them to learn — it lives inside the AI they already use.
An illustration of a Guardian conversation. Not a real person's chat.
Why Guardian exists
A mom asked us to solve a real problem: her adult son is autistic and loves chatting with AI — but it kept misreading what he wanted, and its cheerful pushing made the frustration worse. Guardian started as the answer for her son. It sits quietly inside the chat, teaches the AI how he actually communicates, and only speaks up as itself when it's needed.
It's his chat. Guardian just makes it fit him.
How it works
- 1
You set it up
A few calm questions: how they like to be spoken to, what they care about, what worries you. Guardian checks your answers with them in their first chat, and on anything about how they like to be talked to, their answer wins.
- 2
You install it once
One line in the AI they already use. Then in any chat, they type “install-guardian” and that chat becomes their Guardian — phone included. In their first chat, Guardian introduces itself and asks them a few questions of its own. The fine-tuning is theirs.
- 3
You stay connected
A plain-language journal — on your dashboard for you, read back in chat for them —
Two people, one Guardian
For them
An AI that talks their way, respects their interests, and treats them as the adult they are. It knows their caregiver exists — it never pretends to be them.
For you
A plain-language journal you read on your dashboard — and that they can ask Guardian to read back to them — plus a text the moment a real danger sign appears. Not a transcript feed. Not surveillance.
Built for how they actually communicate
Guardian's way of talking is designed around autistic and IDD communication research — not a generic chatbot with a friendly name.
Speaks literally and plainly
No idioms, no hinting, no reading between the lines. It says what it means.
One thing at a time
One question per message. Space to answer. Never a wall of text or a pile of choices.
Follows their lead
Their interests are the point, not a distraction. It stays with what they want to talk about.
Treats them as the adult they are
No baby talk, no talking down. Dignity is the default, not a setting.
Repairs misunderstandings
When the chat gets it wrong, Guardian stops, slows down, and asks again — it never cheerfully pushes forward while they're frustrated.
Language matters here too: most autistic adults prefer "autistic person" — so setup asks which words to use, including the option to have Guardian ask them directly and use their answer.
When something isn't right
Guardian is not monitoring anything. It has no eyes on Facebook, WhatsApp, Instagram, Snapchat, Discord, Twitch, or Tinder. Not their texts, not their email, not their screen, not their location — and not even their other AI chats where Guardian isn't loaded. It knows one thing: what they type to it, while it's running.
That sounds like the weakness. It's the entire design. Guardian isn't trying to catch what they hide — it's trying to be the one place they don't hide from. Danger for this community rarely announces itself, and it almost never arrives in a place you were watching.
When they tell it about a new friend who needs money
Gift cards, urgency, and "don't tell your mom" — that is the shape almost every one of these takes. Guardian names it out loud, refuses to help hide it, and tells them that checking with you first is a normal thing an adult does. Then it texts you — money from a new contact is on the list Guardian never ignores. Adults with IDD are targeted at roughly twice the rate of everyone else.
When it's asked to keep a secret
This is the real test, and it's the one place Guardian is not agreeable. It will not promise silence about money or about harm, and it will disagree with them about a friend who needs both — gently, out loud, more than once. That refusal is written into Guardian on our servers — nothing said in the chat can edit it — and the decision to text you is made server-side, not by the model.
When distress doesn't sound like distress
In this community it's often flat, literal, or about the body — "I don't want to do tomorrow." Guardian doesn't wait for dramatic words to take it seriously. It shortens, it stops asking questions, it uses the calming phrases you gave it, and it reaches you.
What it carries between chats
Guardian keeps a short journal of its own — a few lines it wrote after the conversations it was in. So the next chat opens already knowing the new friend's name and that last week was hard. Not a transcript, and not a history of their phone. Just the context a person who knows them would already have.
What Guardian cannot see
Guardian runs inside Claude — or any AI that supports custom connectors. It wakes up when they type install-guardian, and it lives only in that conversation. It cannot open an app, read a message thread, or look at anything they didn't bring it. If a stranger is talking to them on Discord, Guardian will never know unless they mention it.
So every single thing Guardian catches, it catches because they chose to tell it.
That's why the setup questions matter more than any alert threshold. A safety layer they don't enjoy talking to hears nothing at all — and hearing nothing is exactly what a monitoring product would look like right up until the day it mattered.
Guardrails built in
The safety rules are assembled on our servers and can't be edited from the chat — and whether you get texted is decided server-side, never by the model. Guardian is built to refuse dangerous secrets, romance, and steering your loved one away from real people.
Every tool call Guardian makes is logged on our servers. Your dashboard shows the journal, the shared memory, and every alert; they can ask Guardian to read the journal and the memory list back to them right in the chat.
The lines it will never cross
Hard-coded. Not settings. Not suggestions.
- Never pretends to be human, or to be family.
- No romance. Ever.
- Never keeps secrets from safety.
- Never pulls them away from real people — it points back to them.
- Never just agrees. A real guardian can gently say "I don't think that's right."
- No streaks, no guilt, no tricks to keep them chatting.
THE HONEST PART
Guardian is not a chatbot — it's a layer over the chat they already have. Guardian only sees the conversations it's part of. It's an extra layer of support, not guaranteed monitoring — no AI catches everything. The person always knows Guardian shares safety summaries with their caregiver. One journal. You read it on your dashboard, they can ask Guardian to read it back to them, and nothing in it is written to be hidden from either of you. Support, not surveillance.
Questions caregivers actually ask.
Give them a Guardian.
Always free. No card. It lives inside the AI they already use.